2 * Copyright 1995-2016 The OpenSSL Project Authors. All Rights Reserved.
4 * Licensed under the OpenSSL license (the "License"). You may not use
5 * this file except in compliance with the License. You can obtain a copy
6 * in the file LICENSE in the source distribution or at
7 * https://www.openssl.org/source/license.html
11 #include <openssl/crypto.h>
12 #include "internal/cryptlib.h"
15 #if defined(BN_LLONG) || defined(BN_UMULT_HIGH)
17 BN_ULONG bn_mul_add_words(BN_ULONG *rp, const BN_ULONG *ap, int num,
26 # ifndef OPENSSL_SMALL_FOOTPRINT
28 mul_add(rp[0], ap[0], w, c1);
29 mul_add(rp[1], ap[1], w, c1);
30 mul_add(rp[2], ap[2], w, c1);
31 mul_add(rp[3], ap[3], w, c1);
38 mul_add(rp[0], ap[0], w, c1);
47 BN_ULONG bn_mul_words(BN_ULONG *rp, const BN_ULONG *ap, int num, BN_ULONG w)
55 # ifndef OPENSSL_SMALL_FOOTPRINT
57 mul(rp[0], ap[0], w, c1);
58 mul(rp[1], ap[1], w, c1);
59 mul(rp[2], ap[2], w, c1);
60 mul(rp[3], ap[3], w, c1);
67 mul(rp[0], ap[0], w, c1);
75 void bn_sqr_words(BN_ULONG *r, const BN_ULONG *a, int n)
81 # ifndef OPENSSL_SMALL_FOOTPRINT
83 sqr(r[0], r[1], a[0]);
84 sqr(r[2], r[3], a[1]);
85 sqr(r[4], r[5], a[2]);
86 sqr(r[6], r[7], a[3]);
93 sqr(r[0], r[1], a[0]);
100 #else /* !(defined(BN_LLONG) ||
101 * defined(BN_UMULT_HIGH)) */
103 BN_ULONG bn_mul_add_words(BN_ULONG *rp, const BN_ULONG *ap, int num,
111 return ((BN_ULONG)0);
116 # ifndef OPENSSL_SMALL_FOOTPRINT
118 mul_add(rp[0], ap[0], bl, bh, c);
119 mul_add(rp[1], ap[1], bl, bh, c);
120 mul_add(rp[2], ap[2], bl, bh, c);
121 mul_add(rp[3], ap[3], bl, bh, c);
128 mul_add(rp[0], ap[0], bl, bh, c);
136 BN_ULONG bn_mul_words(BN_ULONG *rp, const BN_ULONG *ap, int num, BN_ULONG w)
143 return ((BN_ULONG)0);
148 # ifndef OPENSSL_SMALL_FOOTPRINT
150 mul(rp[0], ap[0], bl, bh, carry);
151 mul(rp[1], ap[1], bl, bh, carry);
152 mul(rp[2], ap[2], bl, bh, carry);
153 mul(rp[3], ap[3], bl, bh, carry);
160 mul(rp[0], ap[0], bl, bh, carry);
168 void bn_sqr_words(BN_ULONG *r, const BN_ULONG *a, int n)
174 # ifndef OPENSSL_SMALL_FOOTPRINT
176 sqr64(r[0], r[1], a[0]);
177 sqr64(r[2], r[3], a[1]);
178 sqr64(r[4], r[5], a[2]);
179 sqr64(r[6], r[7], a[3]);
186 sqr64(r[0], r[1], a[0]);
193 #endif /* !(defined(BN_LLONG) ||
194 * defined(BN_UMULT_HIGH)) */
196 #if defined(BN_LLONG) && defined(BN_DIV2W)
198 BN_ULONG bn_div_words(BN_ULONG h, BN_ULONG l, BN_ULONG d)
200 return ((BN_ULONG)(((((BN_ULLONG) h) << BN_BITS2) | l) / (BN_ULLONG) d));
205 /* Divide h,l by d and return the result. */
206 /* I need to test this some more :-( */
207 BN_ULONG bn_div_words(BN_ULONG h, BN_ULONG l, BN_ULONG d)
209 BN_ULONG dh, dl, q, ret = 0, th, tl, t;
215 i = BN_num_bits_word(d);
216 assert((i == BN_BITS2) || (h <= (BN_ULONG)1 << i));
224 h = (h << i) | (l >> (BN_BITS2 - i));
227 dh = (d & BN_MASK2h) >> BN_BITS4;
228 dl = (d & BN_MASK2l);
230 if ((h >> BN_BITS4) == dh)
239 if ((t & BN_MASK2h) ||
240 ((tl) <= ((t << BN_BITS4) | ((l & BN_MASK2h) >> BN_BITS4))))
246 t = (tl >> BN_BITS4);
247 tl = (tl << BN_BITS4) & BN_MASK2h;
263 h = ((h << BN_BITS4) | (l >> BN_BITS4)) & BN_MASK2;
264 l = (l & BN_MASK2l) << BN_BITS4;
269 #endif /* !defined(BN_LLONG) && defined(BN_DIV2W) */
272 BN_ULONG bn_add_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b,
279 return ((BN_ULONG)0);
281 # ifndef OPENSSL_SMALL_FOOTPRINT
283 ll += (BN_ULLONG) a[0] + b[0];
284 r[0] = (BN_ULONG)ll & BN_MASK2;
286 ll += (BN_ULLONG) a[1] + b[1];
287 r[1] = (BN_ULONG)ll & BN_MASK2;
289 ll += (BN_ULLONG) a[2] + b[2];
290 r[2] = (BN_ULONG)ll & BN_MASK2;
292 ll += (BN_ULLONG) a[3] + b[3];
293 r[3] = (BN_ULONG)ll & BN_MASK2;
302 ll += (BN_ULLONG) a[0] + b[0];
303 r[0] = (BN_ULONG)ll & BN_MASK2;
310 return ((BN_ULONG)ll);
312 #else /* !BN_LLONG */
313 BN_ULONG bn_add_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b,
320 return ((BN_ULONG)0);
323 # ifndef OPENSSL_SMALL_FOOTPRINT
326 t = (t + c) & BN_MASK2;
328 l = (t + b[0]) & BN_MASK2;
332 t = (t + c) & BN_MASK2;
334 l = (t + b[1]) & BN_MASK2;
338 t = (t + c) & BN_MASK2;
340 l = (t + b[2]) & BN_MASK2;
344 t = (t + c) & BN_MASK2;
346 l = (t + b[3]) & BN_MASK2;
357 t = (t + c) & BN_MASK2;
359 l = (t + b[0]) & BN_MASK2;
367 return ((BN_ULONG)c);
369 #endif /* !BN_LLONG */
371 BN_ULONG bn_sub_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b,
379 return ((BN_ULONG)0);
381 #ifndef OPENSSL_SMALL_FOOTPRINT
385 r[0] = (t1 - t2 - c) & BN_MASK2;
390 r[1] = (t1 - t2 - c) & BN_MASK2;
395 r[2] = (t1 - t2 - c) & BN_MASK2;
400 r[3] = (t1 - t2 - c) & BN_MASK2;
412 r[0] = (t1 - t2 - c) & BN_MASK2;
423 #if defined(BN_MUL_COMBA) && !defined(OPENSSL_SMALL_FOOTPRINT)
425 # undef bn_mul_comba8
426 # undef bn_mul_comba4
427 # undef bn_sqr_comba8
428 # undef bn_sqr_comba4
430 /* mul_add_c(a,b,c0,c1,c2) -- c+=a*b for three word number c=(c2,c1,c0) */
431 /* mul_add_c2(a,b,c0,c1,c2) -- c+=2*a*b for three word number c=(c2,c1,c0) */
432 /* sqr_add_c(a,i,c0,c1,c2) -- c+=a[i]^2 for three word number c=(c2,c1,c0) */
434 * sqr_add_c2(a,i,c0,c1,c2) -- c+=2*a[i]*a[j] for three word number
440 * Keep in mind that additions to multiplication result can not
441 * overflow, because its high half cannot be all-ones.
443 # define mul_add_c(a,b,c0,c1,c2) do { \
445 BN_ULLONG t = (BN_ULLONG)(a)*(b); \
446 t += c0; /* no carry */ \
447 c0 = (BN_ULONG)Lw(t); \
448 hi = (BN_ULONG)Hw(t); \
449 c1 = (c1+hi)&BN_MASK2; if (c1<hi) c2++; \
452 # define mul_add_c2(a,b,c0,c1,c2) do { \
454 BN_ULLONG t = (BN_ULLONG)(a)*(b); \
455 BN_ULLONG tt = t+c0; /* no carry */ \
456 c0 = (BN_ULONG)Lw(tt); \
457 hi = (BN_ULONG)Hw(tt); \
458 c1 = (c1+hi)&BN_MASK2; if (c1<hi) c2++; \
459 t += c0; /* no carry */ \
460 c0 = (BN_ULONG)Lw(t); \
461 hi = (BN_ULONG)Hw(t); \
462 c1 = (c1+hi)&BN_MASK2; if (c1<hi) c2++; \
465 # define sqr_add_c(a,i,c0,c1,c2) do { \
467 BN_ULLONG t = (BN_ULLONG)a[i]*a[i]; \
468 t += c0; /* no carry */ \
469 c0 = (BN_ULONG)Lw(t); \
470 hi = (BN_ULONG)Hw(t); \
471 c1 = (c1+hi)&BN_MASK2; if (c1<hi) c2++; \
474 # define sqr_add_c2(a,i,j,c0,c1,c2) \
475 mul_add_c2((a)[i],(a)[j],c0,c1,c2)
477 # elif defined(BN_UMULT_LOHI)
479 * Keep in mind that additions to hi can not overflow, because
480 * the high word of a multiplication result cannot be all-ones.
482 # define mul_add_c(a,b,c0,c1,c2) do { \
483 BN_ULONG ta = (a), tb = (b); \
485 BN_UMULT_LOHI(lo,hi,ta,tb); \
486 c0 += lo; hi += (c0<lo)?1:0; \
487 c1 += hi; c2 += (c1<hi)?1:0; \
490 # define mul_add_c2(a,b,c0,c1,c2) do { \
491 BN_ULONG ta = (a), tb = (b); \
492 BN_ULONG lo, hi, tt; \
493 BN_UMULT_LOHI(lo,hi,ta,tb); \
494 c0 += lo; tt = hi+((c0<lo)?1:0); \
495 c1 += tt; c2 += (c1<tt)?1:0; \
496 c0 += lo; hi += (c0<lo)?1:0; \
497 c1 += hi; c2 += (c1<hi)?1:0; \
500 # define sqr_add_c(a,i,c0,c1,c2) do { \
501 BN_ULONG ta = (a)[i]; \
503 BN_UMULT_LOHI(lo,hi,ta,ta); \
504 c0 += lo; hi += (c0<lo)?1:0; \
505 c1 += hi; c2 += (c1<hi)?1:0; \
508 # define sqr_add_c2(a,i,j,c0,c1,c2) \
509 mul_add_c2((a)[i],(a)[j],c0,c1,c2)
511 # elif defined(BN_UMULT_HIGH)
513 * Keep in mind that additions to hi can not overflow, because
514 * the high word of a multiplication result cannot be all-ones.
516 # define mul_add_c(a,b,c0,c1,c2) do { \
517 BN_ULONG ta = (a), tb = (b); \
518 BN_ULONG lo = ta * tb; \
519 BN_ULONG hi = BN_UMULT_HIGH(ta,tb); \
520 c0 += lo; hi += (c0<lo)?1:0; \
521 c1 += hi; c2 += (c1<hi)?1:0; \
524 # define mul_add_c2(a,b,c0,c1,c2) do { \
525 BN_ULONG ta = (a), tb = (b), tt; \
526 BN_ULONG lo = ta * tb; \
527 BN_ULONG hi = BN_UMULT_HIGH(ta,tb); \
528 c0 += lo; tt = hi + ((c0<lo)?1:0); \
529 c1 += tt; c2 += (c1<tt)?1:0; \
530 c0 += lo; hi += (c0<lo)?1:0; \
531 c1 += hi; c2 += (c1<hi)?1:0; \
534 # define sqr_add_c(a,i,c0,c1,c2) do { \
535 BN_ULONG ta = (a)[i]; \
536 BN_ULONG lo = ta * ta; \
537 BN_ULONG hi = BN_UMULT_HIGH(ta,ta); \
538 c0 += lo; hi += (c0<lo)?1:0; \
539 c1 += hi; c2 += (c1<hi)?1:0; \
542 # define sqr_add_c2(a,i,j,c0,c1,c2) \
543 mul_add_c2((a)[i],(a)[j],c0,c1,c2)
545 # else /* !BN_LLONG */
547 * Keep in mind that additions to hi can not overflow, because
548 * the high word of a multiplication result cannot be all-ones.
550 # define mul_add_c(a,b,c0,c1,c2) do { \
551 BN_ULONG lo = LBITS(a), hi = HBITS(a); \
552 BN_ULONG bl = LBITS(b), bh = HBITS(b); \
553 mul64(lo,hi,bl,bh); \
554 c0 = (c0+lo)&BN_MASK2; if (c0<lo) hi++; \
555 c1 = (c1+hi)&BN_MASK2; if (c1<hi) c2++; \
558 # define mul_add_c2(a,b,c0,c1,c2) do { \
560 BN_ULONG lo = LBITS(a), hi = HBITS(a); \
561 BN_ULONG bl = LBITS(b), bh = HBITS(b); \
562 mul64(lo,hi,bl,bh); \
564 c0 = (c0+lo)&BN_MASK2; if (c0<lo) tt++; \
565 c1 = (c1+tt)&BN_MASK2; if (c1<tt) c2++; \
566 c0 = (c0+lo)&BN_MASK2; if (c0<lo) hi++; \
567 c1 = (c1+hi)&BN_MASK2; if (c1<hi) c2++; \
570 # define sqr_add_c(a,i,c0,c1,c2) do { \
572 sqr64(lo,hi,(a)[i]); \
573 c0 = (c0+lo)&BN_MASK2; if (c0<lo) hi++; \
574 c1 = (c1+hi)&BN_MASK2; if (c1<hi) c2++; \
577 # define sqr_add_c2(a,i,j,c0,c1,c2) \
578 mul_add_c2((a)[i],(a)[j],c0,c1,c2)
579 # endif /* !BN_LLONG */
581 void bn_mul_comba8(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b)
588 mul_add_c(a[0], b[0], c1, c2, c3);
591 mul_add_c(a[0], b[1], c2, c3, c1);
592 mul_add_c(a[1], b[0], c2, c3, c1);
595 mul_add_c(a[2], b[0], c3, c1, c2);
596 mul_add_c(a[1], b[1], c3, c1, c2);
597 mul_add_c(a[0], b[2], c3, c1, c2);
600 mul_add_c(a[0], b[3], c1, c2, c3);
601 mul_add_c(a[1], b[2], c1, c2, c3);
602 mul_add_c(a[2], b[1], c1, c2, c3);
603 mul_add_c(a[3], b[0], c1, c2, c3);
606 mul_add_c(a[4], b[0], c2, c3, c1);
607 mul_add_c(a[3], b[1], c2, c3, c1);
608 mul_add_c(a[2], b[2], c2, c3, c1);
609 mul_add_c(a[1], b[3], c2, c3, c1);
610 mul_add_c(a[0], b[4], c2, c3, c1);
613 mul_add_c(a[0], b[5], c3, c1, c2);
614 mul_add_c(a[1], b[4], c3, c1, c2);
615 mul_add_c(a[2], b[3], c3, c1, c2);
616 mul_add_c(a[3], b[2], c3, c1, c2);
617 mul_add_c(a[4], b[1], c3, c1, c2);
618 mul_add_c(a[5], b[0], c3, c1, c2);
621 mul_add_c(a[6], b[0], c1, c2, c3);
622 mul_add_c(a[5], b[1], c1, c2, c3);
623 mul_add_c(a[4], b[2], c1, c2, c3);
624 mul_add_c(a[3], b[3], c1, c2, c3);
625 mul_add_c(a[2], b[4], c1, c2, c3);
626 mul_add_c(a[1], b[5], c1, c2, c3);
627 mul_add_c(a[0], b[6], c1, c2, c3);
630 mul_add_c(a[0], b[7], c2, c3, c1);
631 mul_add_c(a[1], b[6], c2, c3, c1);
632 mul_add_c(a[2], b[5], c2, c3, c1);
633 mul_add_c(a[3], b[4], c2, c3, c1);
634 mul_add_c(a[4], b[3], c2, c3, c1);
635 mul_add_c(a[5], b[2], c2, c3, c1);
636 mul_add_c(a[6], b[1], c2, c3, c1);
637 mul_add_c(a[7], b[0], c2, c3, c1);
640 mul_add_c(a[7], b[1], c3, c1, c2);
641 mul_add_c(a[6], b[2], c3, c1, c2);
642 mul_add_c(a[5], b[3], c3, c1, c2);
643 mul_add_c(a[4], b[4], c3, c1, c2);
644 mul_add_c(a[3], b[5], c3, c1, c2);
645 mul_add_c(a[2], b[6], c3, c1, c2);
646 mul_add_c(a[1], b[7], c3, c1, c2);
649 mul_add_c(a[2], b[7], c1, c2, c3);
650 mul_add_c(a[3], b[6], c1, c2, c3);
651 mul_add_c(a[4], b[5], c1, c2, c3);
652 mul_add_c(a[5], b[4], c1, c2, c3);
653 mul_add_c(a[6], b[3], c1, c2, c3);
654 mul_add_c(a[7], b[2], c1, c2, c3);
657 mul_add_c(a[7], b[3], c2, c3, c1);
658 mul_add_c(a[6], b[4], c2, c3, c1);
659 mul_add_c(a[5], b[5], c2, c3, c1);
660 mul_add_c(a[4], b[6], c2, c3, c1);
661 mul_add_c(a[3], b[7], c2, c3, c1);
664 mul_add_c(a[4], b[7], c3, c1, c2);
665 mul_add_c(a[5], b[6], c3, c1, c2);
666 mul_add_c(a[6], b[5], c3, c1, c2);
667 mul_add_c(a[7], b[4], c3, c1, c2);
670 mul_add_c(a[7], b[5], c1, c2, c3);
671 mul_add_c(a[6], b[6], c1, c2, c3);
672 mul_add_c(a[5], b[7], c1, c2, c3);
675 mul_add_c(a[6], b[7], c2, c3, c1);
676 mul_add_c(a[7], b[6], c2, c3, c1);
679 mul_add_c(a[7], b[7], c3, c1, c2);
684 void bn_mul_comba4(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b)
691 mul_add_c(a[0], b[0], c1, c2, c3);
694 mul_add_c(a[0], b[1], c2, c3, c1);
695 mul_add_c(a[1], b[0], c2, c3, c1);
698 mul_add_c(a[2], b[0], c3, c1, c2);
699 mul_add_c(a[1], b[1], c3, c1, c2);
700 mul_add_c(a[0], b[2], c3, c1, c2);
703 mul_add_c(a[0], b[3], c1, c2, c3);
704 mul_add_c(a[1], b[2], c1, c2, c3);
705 mul_add_c(a[2], b[1], c1, c2, c3);
706 mul_add_c(a[3], b[0], c1, c2, c3);
709 mul_add_c(a[3], b[1], c2, c3, c1);
710 mul_add_c(a[2], b[2], c2, c3, c1);
711 mul_add_c(a[1], b[3], c2, c3, c1);
714 mul_add_c(a[2], b[3], c3, c1, c2);
715 mul_add_c(a[3], b[2], c3, c1, c2);
718 mul_add_c(a[3], b[3], c1, c2, c3);
723 void bn_sqr_comba8(BN_ULONG *r, const BN_ULONG *a)
730 sqr_add_c(a, 0, c1, c2, c3);
733 sqr_add_c2(a, 1, 0, c2, c3, c1);
736 sqr_add_c(a, 1, c3, c1, c2);
737 sqr_add_c2(a, 2, 0, c3, c1, c2);
740 sqr_add_c2(a, 3, 0, c1, c2, c3);
741 sqr_add_c2(a, 2, 1, c1, c2, c3);
744 sqr_add_c(a, 2, c2, c3, c1);
745 sqr_add_c2(a, 3, 1, c2, c3, c1);
746 sqr_add_c2(a, 4, 0, c2, c3, c1);
749 sqr_add_c2(a, 5, 0, c3, c1, c2);
750 sqr_add_c2(a, 4, 1, c3, c1, c2);
751 sqr_add_c2(a, 3, 2, c3, c1, c2);
754 sqr_add_c(a, 3, c1, c2, c3);
755 sqr_add_c2(a, 4, 2, c1, c2, c3);
756 sqr_add_c2(a, 5, 1, c1, c2, c3);
757 sqr_add_c2(a, 6, 0, c1, c2, c3);
760 sqr_add_c2(a, 7, 0, c2, c3, c1);
761 sqr_add_c2(a, 6, 1, c2, c3, c1);
762 sqr_add_c2(a, 5, 2, c2, c3, c1);
763 sqr_add_c2(a, 4, 3, c2, c3, c1);
766 sqr_add_c(a, 4, c3, c1, c2);
767 sqr_add_c2(a, 5, 3, c3, c1, c2);
768 sqr_add_c2(a, 6, 2, c3, c1, c2);
769 sqr_add_c2(a, 7, 1, c3, c1, c2);
772 sqr_add_c2(a, 7, 2, c1, c2, c3);
773 sqr_add_c2(a, 6, 3, c1, c2, c3);
774 sqr_add_c2(a, 5, 4, c1, c2, c3);
777 sqr_add_c(a, 5, c2, c3, c1);
778 sqr_add_c2(a, 6, 4, c2, c3, c1);
779 sqr_add_c2(a, 7, 3, c2, c3, c1);
782 sqr_add_c2(a, 7, 4, c3, c1, c2);
783 sqr_add_c2(a, 6, 5, c3, c1, c2);
786 sqr_add_c(a, 6, c1, c2, c3);
787 sqr_add_c2(a, 7, 5, c1, c2, c3);
790 sqr_add_c2(a, 7, 6, c2, c3, c1);
793 sqr_add_c(a, 7, c3, c1, c2);
798 void bn_sqr_comba4(BN_ULONG *r, const BN_ULONG *a)
805 sqr_add_c(a, 0, c1, c2, c3);
808 sqr_add_c2(a, 1, 0, c2, c3, c1);
811 sqr_add_c(a, 1, c3, c1, c2);
812 sqr_add_c2(a, 2, 0, c3, c1, c2);
815 sqr_add_c2(a, 3, 0, c1, c2, c3);
816 sqr_add_c2(a, 2, 1, c1, c2, c3);
819 sqr_add_c(a, 2, c2, c3, c1);
820 sqr_add_c2(a, 3, 1, c2, c3, c1);
823 sqr_add_c2(a, 3, 2, c3, c1, c2);
826 sqr_add_c(a, 3, c1, c2, c3);
831 # ifdef OPENSSL_NO_ASM
832 # ifdef OPENSSL_BN_ASM_MONT
835 * This is essentially reference implementation, which may or may not
836 * result in performance improvement. E.g. on IA-32 this routine was
837 * observed to give 40% faster rsa1024 private key operations and 10%
838 * faster rsa4096 ones, while on AMD64 it improves rsa1024 sign only
839 * by 10% and *worsens* rsa4096 sign by 15%. Once again, it's a
840 * reference implementation, one to be used as starting point for
841 * platform-specific assembler. Mentioned numbers apply to compiler
842 * generated code compiled with and without -DOPENSSL_BN_ASM_MONT and
843 * can vary not only from platform to platform, but even for compiler
844 * versions. Assembler vs. assembler improvement coefficients can
845 * [and are known to] differ and are to be documented elsewhere.
847 int bn_mul_mont(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,
848 const BN_ULONG *np, const BN_ULONG *n0p, int num)
850 BN_ULONG c0, c1, ml, *tp, n0;
854 volatile BN_ULONG *vp;
857 # if 0 /* template for platform-specific
860 return bn_sqr_mont(rp, ap, np, n0p, num);
862 vp = tp = alloca((num + 2) * sizeof(BN_ULONG));
871 for (j = 0; j < num; ++j)
872 mul(tp[j], ap[j], ml, mh, c0);
874 for (j = 0; j < num; ++j)
875 mul(tp[j], ap[j], ml, c0);
882 for (i = 0; i < num; i++) {
888 for (j = 0; j < num; ++j)
889 mul_add(tp[j], ap[j], ml, mh, c0);
891 for (j = 0; j < num; ++j)
892 mul_add(tp[j], ap[j], ml, c0);
894 c1 = (tp[num] + c0) & BN_MASK2;
896 tp[num + 1] = (c1 < c0 ? 1 : 0);
899 ml = (c1 * n0) & BN_MASK2;
904 mul_add(c1, np[0], ml, mh, c0);
906 mul_add(c1, ml, np[0], c0);
908 for (j = 1; j < num; j++) {
911 mul_add(c1, np[j], ml, mh, c0);
913 mul_add(c1, ml, np[j], c0);
915 tp[j - 1] = c1 & BN_MASK2;
917 c1 = (tp[num] + c0) & BN_MASK2;
919 tp[num] = tp[num + 1] + (c1 < c0 ? 1 : 0);
922 if (tp[num] != 0 || tp[num - 1] >= np[num - 1]) {
923 c0 = bn_sub_words(rp, tp, np, num);
924 if (tp[num] != 0 || c0 == 0) {
925 for (i = 0; i < num + 2; i++)
930 for (i = 0; i < num; i++)
931 rp[i] = tp[i], vp[i] = 0;
938 * Return value of 0 indicates that multiplication/convolution was not
939 * performed to signal the caller to fall down to alternative/original
942 int bn_mul_mont(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,
943 const BN_ULONG *np, const BN_ULONG *n0, int num)
947 # endif /* OPENSSL_BN_ASM_MONT */
950 #else /* !BN_MUL_COMBA */
952 /* hmm... is it faster just to do a multiply? */
953 # undef bn_sqr_comba4
954 # undef bn_sqr_comba8
955 void bn_sqr_comba4(BN_ULONG *r, const BN_ULONG *a)
958 bn_sqr_normal(r, a, 4, t);
961 void bn_sqr_comba8(BN_ULONG *r, const BN_ULONG *a)
964 bn_sqr_normal(r, a, 8, t);
967 void bn_mul_comba4(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b)
969 r[4] = bn_mul_words(&(r[0]), a, 4, b[0]);
970 r[5] = bn_mul_add_words(&(r[1]), a, 4, b[1]);
971 r[6] = bn_mul_add_words(&(r[2]), a, 4, b[2]);
972 r[7] = bn_mul_add_words(&(r[3]), a, 4, b[3]);
975 void bn_mul_comba8(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b)
977 r[8] = bn_mul_words(&(r[0]), a, 8, b[0]);
978 r[9] = bn_mul_add_words(&(r[1]), a, 8, b[1]);
979 r[10] = bn_mul_add_words(&(r[2]), a, 8, b[2]);
980 r[11] = bn_mul_add_words(&(r[3]), a, 8, b[3]);
981 r[12] = bn_mul_add_words(&(r[4]), a, 8, b[4]);
982 r[13] = bn_mul_add_words(&(r[5]), a, 8, b[5]);
983 r[14] = bn_mul_add_words(&(r[6]), a, 8, b[6]);
984 r[15] = bn_mul_add_words(&(r[7]), a, 8, b[7]);
987 # ifdef OPENSSL_NO_ASM
988 # ifdef OPENSSL_BN_ASM_MONT
990 int bn_mul_mont(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,
991 const BN_ULONG *np, const BN_ULONG *n0p, int num)
993 BN_ULONG c0, c1, *tp, n0 = *n0p;
994 volatile BN_ULONG *vp;
997 vp = tp = alloca((num + 2) * sizeof(BN_ULONG));
999 for (i = 0; i <= num; i++)
1002 for (i = 0; i < num; i++) {
1003 c0 = bn_mul_add_words(tp, ap, num, bp[i]);
1004 c1 = (tp[num] + c0) & BN_MASK2;
1006 tp[num + 1] = (c1 < c0 ? 1 : 0);
1008 c0 = bn_mul_add_words(tp, np, num, tp[0] * n0);
1009 c1 = (tp[num] + c0) & BN_MASK2;
1011 tp[num + 1] += (c1 < c0 ? 1 : 0);
1012 for (j = 0; j <= num; j++)
1016 if (tp[num] != 0 || tp[num - 1] >= np[num - 1]) {
1017 c0 = bn_sub_words(rp, tp, np, num);
1018 if (tp[num] != 0 || c0 == 0) {
1019 for (i = 0; i < num + 2; i++)
1024 for (i = 0; i < num; i++)
1025 rp[i] = tp[i], vp[i] = 0;
1031 int bn_mul_mont(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,
1032 const BN_ULONG *np, const BN_ULONG *n0, int num)
1036 # endif /* OPENSSL_BN_ASM_MONT */
1039 #endif /* !BN_MUL_COMBA */